Most IoT devices run on microcontroller units (MCUs), while most machine learning happens on CPUs. One of the most cutting-edge innovations in AI is the ability to run models on constrained devices. In the past, AI was limited to large computers with traditional operating systems such as Windows or Linux. Now, small devices can execute machine learning models with technologies such as ONYX and TensorFlow Lite. These constrained devices are low cost, can use machine learning without an internet connection, and can save dramatically on cloud costs.

Many IoT projects fail due to high cloud costs. IoT devices are often sold for a fixed price without a reoccurring subscription model. They then incur high cloud costs by performing machine learning or analytics. There is no reason this needs to be the case. Even for microcontrollers, the...